thinkhard.SI

Prompt repository Research & learning

Explain why two reports give different numbers

Two sources disagree on a figure, such as a market size, a rate or a survey result, and you need to know why before you choose one.

Registry no.
RES-018
Collection
Research & learning
Laid out for
Any chat model
Score
10 out of 10
Blanks to fill
4
Added

Research & learningAny chat

The prompt

Score 10 of 10, Excellent

Use it when Two sources disagree on a figure, such as a market size, a rate or a survey result, and you need to know why before you choose one.

  • {{figure}}
  • {{use_and_audience}}
  • {{source_a}}
  • {{source_b}}
Full prompt and why it works
Prompt
The figure in question: {{figure}}
What I need it for, and who will see it: {{use_and_audience}}

Source A (name, date, and the relevant text, including any notes on method, definitions or sample):
"""
{{source_a}}
"""

Source B (the same details):
"""
{{source_b}}
"""

Task:
Explain why these two sources give different numbers for the same thing, so that I can choose which to use for {{use_and_audience}}, or report both honestly. Most disagreements come from differences in what was measured rather than from one source being wrong, so check those first.

Format:
1. The two figures side by side, quoted exactly, each with its date and source.
2. A comparison table with the columns Factor, Source A, Source B, Could this explain the gap? Cover the exact definition of what is counted, geography, time period, sample or data source, method (survey, records, model or estimate), units and currency, and who produced or paid for it.
3. Most likely explanation: one short paragraph ranking the differences by how much of the gap each could explain. Where simple arithmetic from the texts shows it, include it; for instance, one figure in euros and one in dollars, or one including a segment the other leaves out.
4. Which to use: a recommendation for my purpose, or one sentence that reports both fairly.

A good explanation lets me say in one sentence why the figures differ and defend the one I choose.

Example of one table row's form only: "Time period | Calendar year 2023 | 12 months to June 2024 | Partly: B includes six later months."

Requirements:
- Under 450 words, not counting the table.
- Use only the two texts. Where one does not state something, write "Not stated" in the table instead of assuming, since a missing definition is itself a reason to trust a figure less.
- Do not average the two numbers or put forward a third figure as the answer. Arithmetic that explains the gap is fine when each step uses numbers from the texts.

If the texts do not contain enough to explain the gap, say so plainly and list the questions to put to each publisher.

Variables

{{figure}}
What both sources measure, for example "number of small businesses in Texas" or "annual staff turnover in retail".
{{use_and_audience}}
Where the figure will go and who sees it, for example "a market sizing slide for our investors".
{{source_a}}
The first source's name, date and the text around the figure, including any methods or definitions section.
{{source_b}}
The second source, with the same details.

Why it works

  • Format. A factor-by-factor table of definition, period, sample, method and units finds the real reason for most disagreements, which is that the two sources counted different things.
  • What to avoid. It never averages the two numbers or makes up a third, so whatever you report can be traced to a source.
  • If unsure. A definition a source does not state goes in the table as "Not stated", and where the texts cannot explain the gap, you get questions for each publisher.

How to use it

  1. Press Fill in variables and type your details into each blank. What you type stays in this browser.
  2. Copy the filled prompt into the AI tool you already use, such as ChatGPT, Claude or Gemini, or open it in the Grader or the Prompt Tester here.
  3. Before you rely on it, try it on three or four real examples. The Prompt Tester runs them all at once and shows which ones go wrong.
Use it from code

Get this prompt as JSON from the free REST API, with your own text in the blanks, or let an AI assistant fetch it through the MCP server. No key.

curl -s https://thinkhard.si/api/v1/prompts/reports-with-different-numbers

With the MCP server added to Claude, ChatGPT, Cursor or VS Code, call get_prompt with {"prompt": "reports-with-different-numbers"}, or just ask for it by name.